BGP config WAS RE: [c-nsp] internet redundancy [7:99340]

[Howard C. Berkowitz]

At 11:49 AM -0700 5/19/05, David Barak wrote:
>in-line.
>
>--- Vikas Sharma <vikassharmas at gmail.com> wrote:
>
>>  Hi,
>>   1) You don't need to pay for using ASN as this is
>>  default BGP attribute.
>
>the fact than an ASN is a default attribute has
>nothing whatsoever to do with payment.  In any case,
>Tom will need to contact a RIR (ARIN, RIPE, etc) and
>obtain a public ASN, and for this, he'll have to pay a
>nominal annual fee.

$500 per year at ARIN when I last looked -- www.arin.net for the most 
recent information, of course.  ASNs in the global Internet have to 
be unique, so there have to be AS registries to manage them -- and 
that costs something.

ARIN, as opposed to RIPE and I believe APNIC, does not require you to 
record your routing policy, written in RPSL, in their routing 
registry. Nevertheless, it is very good practice to do so.

>
>
>>  2) It is better if u use IBGP internally because
>>  IBGP support all attribute
>>  of BGP and one should avoid redistribution as
>  > possible.

A safe guideline is that you should never, in real-world Internet 
routing, redistribute your IGP into BGP, or BGP into your IGP.  When 
you learn enough about global routing to know the hazards, you may 
make some very careful exceptions -- and will very, very rarely want 
do so.


>  > 3) You need your ASN only when u want to differ from
>  > ISP policies otherwise
>>  u don't require ASN.
>
>Tom is connecting to multiple ISPs - therefore, BGP is
>the appropriate solution.

Correct.

>
>
>>  4) To filter you can use
>>   a) Prefix filtering
>>   b) AS_path filtering
>>   c) Route map fltering
>>   d) Community filtering
>
>The most effective type of filtering in this case is
>done on the provider's routers - just ask the provider
>for their networks and their customer networks, and
>all will be well.

What problem are you solving with the filtering?  Each of those 
methods, except route map filtering (unless you mean something else 
than I think), are normally written inside route maps. There are 
three different mechanisms there because they are useful for at least 
three different things.

>
>>   beside this u need to ask ur SP's to broadcast
>>  other SP's route in their
>>  network or u may not need this if BGP
>  > synchronization is enable in all SPs.

If you are using provider-assigned address space, the assigning 
provider MUST advertise both its aggregate and the more-specific. The 
other providers also MUST advertise that more specific, and will 
usually want written confirmation from the assigning ISP that it's OK 
to do so. All three should add that advertisement to their published 
routing policy.

>
>No service provider worth their salt uses BGP
>synchronization.

AFAIK, Juniper, NextHop, Nortel, and Ericsson don't even support 
synchronization.

>If they did, they would only
>advertise routes which appeared in their IGP!  Rather,
>if Tom's ISPs are all cooperative, then they will
>permit him to originate address space, and will carry
>those announcements rather than aggregating them. 
>
>>   Chad :- here internet router and core router are
>>  different. so no need to
>>  think abt internet link go down as these are
>>  seperate router. Redistribution
>>  of default route is a good option with network
>>  0.0.0.0 <http://0.0.0.0> in
>>  bgp, but if that link is down there will not be any
>>  connectivity. You can
>>  also use MED attribute to specify the best route.
>
>ick!  Redistribution of a default route is an ugly
>solution: the better approach is to have the edge
>routers all speak iBGP with each other and the core
>routers which make decisions.  The core routers are
>the default gateways for Tom's internal network, and
>once data reaches the core, it follows a BGP path
>outward.
>
Agreed.